{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/arxiv-2605-21999","title":"Toward Understanding Adversarial Distillation: Why Robust Teachers Fail","arxiv_id":"2605.21999","date":"2026-05-21","proceeding":"ICML","authors":["Hongsin Lee","Hye Won Chung"],"abstract":"Adversarial Distillation aims to enhance student robustness by guiding the student with a robust teacher's soft labels within the min-max adversarial training framework, yet its success is notoriously inconsistent: a more robust teacher often fails to improve, or even harms, the student's robust generalization. In this paper, we identify a key mechanism of this teacher dependency: the misalignment between the teacher's supervisory confidence and the student's representational limitations on a consistent subset of training data -- the Robustly Unlearnable Set. We present a theoretical framework analyzing the feature learning dynamics of a two-layer neural network, demonstrating that this mismatch creates a dichotomy in distillation outcomes. We prove that when a teacher provides confident supervision on unlearnable samples, it compels the student to memorize spurious noise patterns that eventually overpower the learned robust signal, thereby driving robust overfitting. Conversely, a teacher that exhibits high uncertainty on these samples effectively suppresses noise memorization, allowing the student to rely solely on the learnable signal for robust generalization. We empirically validate our theory across both synthetic simulations and real-image classification datasets, confirming that robust overfitting is driven by the teacher's interaction with unlearnable samples. Finally, we demonstrate that a teacher's predictive entropy on unlearnable samples serves as a strong indicator of student robustness, validating our theoretical framework and offering a principled guideline for robust teacher selection.","url_abs":"https://arxiv.org/abs/2605.21999","url_pdf":"https://arxiv.org/pdf/2605.21999","source":{"archive":null,"snapshot":"2025-07-28","note":"not in the Papers with Code archive (frozen at the snapshot)","row_kind":"graph","title_abstract_authors_date":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)"},"code_links":[],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2605.21999","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2605.21999"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"mentioned_in_github":null,"is_official":null,"provenance":"deterministic:regex_extraction","mentioned_in_paper":null,"url":"https://github.com/HongsinLee/why-robust-teachers-fail","reach":null}],"summary":{"ran":8,"unverified":2},"by_repo_kind":{"found_in_text":{"samples":10,"ran":8,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":10,"samples":[{"code_sha256_prefix":"a648902a7268d3fd","entry":"PGD","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/attacks.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/attacks.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"a648902a7268d3fd"}},{"code_sha256_prefix":"1de719bcbbc0bd3c","entry":"TRADES","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/attacks.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/attacks.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"1de719bcbbc0bd3c"}},{"code_sha256_prefix":"f5088fe40e624080","entry":"pResNet18","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/cifar100_models/preresnet.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/cifar100_models/preresnet.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"f5088fe40e624080"}},{"code_sha256_prefix":"fd5e91ae30144db3","entry":"pResNet34","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/cifar100_models/preresnet.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/cifar100_models/preresnet.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"fd5e91ae30144db3"}},{"code_sha256_prefix":"c26044835e29342e","entry":"pResNet50","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/cifar100_models/preresnet.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/cifar100_models/preresnet.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"c26044835e29342e"}},{"code_sha256_prefix":"804dc4361a30432d","entry":"resnet18","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/cifar100_models/resnet.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/cifar100_models/resnet.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"804dc4361a30432d"}},{"code_sha256_prefix":"47e8b2f8a9d376b2","entry":"rslad_inner_loss","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/attacks.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/attacks.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"47e8b2f8a9d376b2"}},{"code_sha256_prefix":"95a525f1ac88c43c","entry":"wideresnet","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/cifar10_models/wideresnet.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/cifar10_models/wideresnet.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"95a525f1ac88c43c"}},{"code_sha256_prefix":"6e78bd6eb54f58a0","entry":"svhn_wideresnetwithswish","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/cifar100_models/wideresnetswish_svhn.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/cifar100_models/wideresnetswish_svhn.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"6e78bd6eb54f58a0"}},{"code_sha256_prefix":"414b9f77b95e1b72","entry":"ti_wideresnetwithswish","repo":"HongsinLee/why-robust-teachers-fail","repo_kind":"found_in_text","path":"real_exp/cifar100_models/wideresnetswish.py","file_url":"https://github.com/HongsinLee/why-robust-teachers-fail/blob/HEAD/real_exp/cifar100_models/wideresnetswish.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"414b9f77b95e1b72"}}]},"arxiv_metadata":{"licence":"arXiv metadata, CC0 1.0 (https://info.arxiv.org/help/license)","fields":["title","abstract","authors","date"],"primary_category":"cs.LG","source":"arxiv_2026.jsonl"},"syntology_extracted_results":null}